This plugin simulates different types of noise on a sequences, with customizable characteristics. This could be useful to test and validate the denoising and restoration algorithms that are implemented in Icy. Currently, three noise models are implemented:
- white additive Gaussian noise,
- Poisson noise (i.e. photon counting noise),
- salt & pepper noise.
How each of these models relates the input clean sequence A and the output noisy sequence B is described below.
The features provided by this plugin can be accessed:
- directly through the GUI (using a EzPlug interface),
- through the protocol editor provided by the Protocols plugin,
White additive Gaussian noise
1 parameter: sigma >= 0 (standard deviation of the Gaussian random variables).
This noise model enforces B = A + n, where n is a random sequence such that the samples n(x,y,z,t,c) are random independant variables following a Gaussian probability distribution of mean 0 and variance sigma^2.
In this model, each output sample B(x,y,z,t,c) is generated from a Poisson random distribution of intensity A(x,y,z,t,c), which is supposed to be >=0. If A(x,y,z,t,c) < 0, then B(x,y,z,t,c) is set to NaN.
Salt & pepper noise
- intensity, that must satisfy 0 <= intensity <= 1,
- lowerBound and upperBound, with lowerBound <= upperBound.
In this model, each output sample B(x,y,z,t,c):
- either is let unchanged (i.e. is set to A(x,y,z,t,c)), with probability intensity,
- or takes a value that is selected in a random uniform manner between lowerBound and upperBound, with probability 1-intensity.